Turn Slack conversations and Linear/Jira tickets into reviewed, tested, merge-ready pull requests. Bring Claude Code or Codex on your own AI accounts. AOSplane adds the planning, code review and QA around them, caps what each task can spend, and records every step. You decide what gets merged.
Sessionacme/web-appExample
Streamlined onboarding flow
Claude CodeOn it, in a private sandbox.
Keep email login working for existing accounts.
Claude CodeDone. Email login still works, and I added a test for it.
Draft pull request#2183Proven
Yours to review and merge.
Runs the coding agents your team already uses
Takes work from where your team already works
Launch video
0:00 / 1:3001 · The problem
The console
Real screens from the AOSplane console. Only the names are samples.
Describe what done looks like, pick the repository and the agent, and choose when it starts and whether it asks you first.
Download
Sign in with your AOSplane account and pick up the same sessions you see on the web.
Desktop · macOS
Your sessions in their own window. Follow each one live, answer what an agent asks for, and see on the Dock how many are waiting on you.
Universal: Apple Silicon and Intel · macOS 10.15 or later
Mobile · Android
Approve what a run asks for, answer its questions and follow it to the pull request, with a notification when it needs you.
iOS: coming soon
The rest of the team
Start a task with Plan first and it runs the way a good team works: a product manager plans it, you approve the plan, your agent builds it, and review and QA check the work before a draft pull request opens.
A change asked for or a failed check goes back to your agent, up to five tries
Turns a problem into a plan you approve before an agent writes any code.
Reads each pull request and comments on what should change.
Runs your repository’s checks and sends the work back until they pass.
Guardrails
Four simple controls make it safe to hand real work to agents, even when nobody is watching.
Spending limits
Set limits for the whole company, for each project and for each agent. Before a task starts, AOSplane checks that all three can cover it, and the lowest limit always wins.
CompanyProjectAgent
Spend, as it happens
Example62% of this project’s monthly limit
Every task’s cost shows up in the console, and counts toward each of its limits.
Every task runs in a fresh, isolated sandbox in the cloud, kept apart from your other tasks and from other customers.
When an agent wants to do something risky, like change a production database, it stops and waits. You approve or deny it from the console.
Each task keeps a step-by-step history: what it was asked, what it ran, and who approved what. Every entry is checked as it’s written, so changes can’t slip in unnoticed.
How it works
Your team keeps working the way it does today. Agents pick up the tasks you hand them and send back changes you can review.
01
Link your own AI account and your GitHub repositories. Model costs go straight to your AI provider, with no markup.
02
Write it in the console, start it from a Linear or Jira issue, hand an issue to AOSplane inside Linear, or send it over the API. Choose which agent and model takes it.
03
The agent works in its own sandbox in the cloud. It can’t push to your repository, so every change comes back as a pull request for you to review.
Tasks
3 running $3.70 today
Add usage-based billing to the API
Claude CodeWriting code
$0.84
Fix the flaky checkout test
CodexRunning tests
$0.31
Migrate the orders table
CodexNeeds you
$0.12
Upgrade the Postgres client
Claude CodePR #482 ready
$0.42
Use cases
Begin with the backlog and the red builds, then hand off whole features once you’ve seen the plan, the review and the checks do their part. Six places to start:
Start a session from any Linear or Jira issue, or hand the issue to AOSplane inside Linear. Each one runs in its own sandbox, side by side with the rest.
Comes backA pull request for each issue
Write the problem, not the plan. The product manager turns it into a spec with acceptance criteria; once you approve it, your agent builds until QA proves every one.
Comes backA draft pull request with QA’s result per criterion
Point your agent at a failing test or a flaky check. After each try QA runs the repository’s checks, and a failure goes back with the command and the end of its output.
Comes backA draft pull request with every check passing
Code review reads pull requests as they open and change, your team’s included, with your AGENTS.md as its conventions. Docs-only and version-bump changes are skipped, free.
Comes backComments on the lines that should change
When an agent wants to do something sensitive, like run a database migration, it stops and asks. You decide in the console, and the task’s record keeps who approved what.
Comes backA pull request, and a record of every step
Project API keys let CI pipelines and scripts start and follow sessions, so a nightly job can hand off the dependency updates. In Slack, mention @AOSplane or use /aos Rolling out
Comes backA pull request your pipeline can follow
Sandboxes
Each task works on a private machine sized for the job. It sleeps when nobody needs it and picks up again when you write.
ShrimpSmall
4 vCPU · 8 GiB
Every plan
TunaMedium
8 vCPU · 16 GiB
Starter and up
SharkLarge
16 vCPU · 32 GiB
Team and up
WhaleExtra large
32 vCPU · 64 GiB
Team and up
A sandbox sleeps after 30 idle minutes by default and wakes when you send its chat a message. Keep awake holds it for up to 24 hours.
Open more chats on a task’s sandbox as tabs, like a browser: each keeps its own conversation, on the same machine.
A sandbox’s page lists its snapshots. Restore one into a new sandbox, in any size your plan offers.
Connections
Hand off work from the tools your team already uses, give agents only the tools you allow, and keep model costs on your own AI account.
Start work where your team already plans it. However it starts, every change comes back as a pull request.
Each project decides which apps its agents may use, with whose account, and which tools need a person’s approval first.
Your own AI account. The provider bills you directly, with no markup. Accounts belong to the organization or to one person, and a known expiry date is flagged 30 days ahead.
Mobile app
Approve what a run asks for, answer its questions, steer a session mid-turn, and follow it to the pull request from your phone.
iOSComing soon
Coming soon
AOSplane SRE reads your alerts, logs and deploys with the access of whoever asks, tells you what it saw and what it only suspects, and turns every fix into a plan an admin approves.
AOSplane SREComing soon
Ask in plain words. It reads Datadog, Sentry, GitHub and PagerDuty with your access and shows the evidence.
Design previews. The names and numbers are samples.
Each question runs with the asker’s own access, so nobody sees more than they already could. In Slack, each person consents first.
A fix comes as a plan for one exact target. It runs only once an admin approves it in AOSplane, never from a chat.
Counting and grouping run inside Datadog. Only summaries and a few sample lines reach the model, with secrets redacted.
File the follow-up in Linear and hand it to your coding agent, which opens the fix as a pull request.
Pricing
Every plan includes monthly usage credit. Model costs go to your own AI account, with no markup.
$0
Free plan
Get your first agents working.
No markup
Your AI account
Model costs go to your own AI account, with no markup.
Per second
Runtime billing
Platform usage is billed by the second while your tasks run.
Usage credit
On every plan
Every plan includes monthly usage credit.
$0/ month
Get your first agents working.
Start free$5 monthly usage credit
$39/ month
Give a small team of agents room to run.
Book a call$25 monthly usage credit
$149/ month
Run agents across your whole team.
Book a call$100 monthly usage credit
Let’s talk
Fit AOSplane to your organization.
Talk to usCustom scale and terms
Platform usage is billed by the second while your tasks run. Free is self-serve; paid plans start with a short call with our team.
Questions
Still curious? Email us and a person will answer.
No. AOSplane runs the coding agents you already use, like Claude Code and Codex, in the cloud. It adds what a team needs around them: a private sandbox for every task, spending limits, approvals for sensitive steps, and a record of everything they did.
AOSplane runs Claude Code, so you keep the agent you like. It adds what a team needs around it: a spec with acceptance criteria you approve before work starts, QA that checks each one before you see the pull request, spending limits checked before a task starts, approvals for sensitive steps, and one record of every step. It does the same for Codex, so a team that uses both runs them one way, on its own accounts.
You choose. When you set up an agent, you pick which coding agent it runs and which model it uses. Matching each task to the best agent automatically is something we are still building.
A new task doesn’t start if its cost would go over any of your limits: the company’s, the project’s or the agent’s. Tasks that are already running are not cancelled by that limit.
Yes. Connect an Anthropic or OpenAI API key, Amazon Bedrock with an AWS role or a bearer token, Claude Platform on AWS, Azure OpenAI, another compatible provider, or a Claude or ChatGPT subscription you already pay for (sign-in for subscriptions is still rolling out). Agents use the account directly, so the provider bills you and its limits apply, with no markup. AOSplane bills separately for platform usage.
An agent can stop and ask for permission before a sensitive step. You see what it wants to do, then approve or deny it. Your answer is saved in the task’s history.
No. Every task runs in its own isolated sandbox, and agents can’t push to your repositories. Their changes come back as a pull request for you to review.
Get started
Connect your AI account, pick a repository and hand off a task. The free plan is ready in minutes.
Read the quickstart